First test of partially generated LBM kernel

A kernel extracted from `lbn_codegen.ipynb` yields ~665 MLUPS compared
to the ~600 MLUPS produced by a manually optimized kernel.

Note that this new kernel currently doesn't handle boundary conditions (but
dropping in a density condition doesn't impact performance).

1 files changed, 0 insertions, 14 deletions

diff --git a/lid_driven_cavity.py b/lid_driven_cavity.py
index cfaf0ba..216ee87 100644
--- a/lid_driven_cavity.py
+++ b/lid_driven_cavity.py
@@ -134,7 +134,6 @@ class D2Q9_BGK_Lattice:
         self.setup_geometry()
 
         self.equilibrilize()
-        #self.setup_anomaly()
 
         self.cl_pop_a = cl.Buffer(self.context, mf.READ_WRITE | mf.USE_HOST_PTR, hostbuf=self.np_pop_a)
         self.cl_pop_b = cl.Buffer(self.context, mf.READ_WRITE | mf.USE_HOST_PTR, hostbuf=self.np_pop_b)
@@ -165,19 +164,6 @@ class D2Q9_BGK_Lattice:
         self.np_pop_b[(1,3,5,7),:] = 1./9.
         self.np_pop_b[4,:] = 4./9.
 
-    def setup_anomaly(self):
-        bubbles = [ [        self.nX//4,        self.nY//4],
-                    [        self.nX//4,self.nY-self.nY//4],
-                    [self.nX-self.nX//4,        self.nY//4],
-                    [self.nX-self.nX//4,self.nY-self.nY//4] ]
-
-        for x in range(0,self.nX-1):
-            for y in range(0,self.nY-1):
-                for [a,b] in bubbles:
-                    if numpy.sqrt((x-a)*(x-a)+(y-b)*(y-b)) < self.nX//10:
-                        self.np_pop_a[:,self.idx(x,y)] = 1./24.
-                        self.np_pop_b[:,self.idx(x,y)] = 1./24.
-
     def build_kernel(self):
         self.program = cl.Program(self.context, Template(kernel).substitute({
             'nX' : self.nX,